Re: small bug fix in ppval

[Jaroslav Hajek]

Shai,

the "l" and "r" options for lookup should be documented in new
lookup's inline documentation (via `help lookup'). If it is not clear,
blame me (and please report what is broken). Shortly, "l" means extend
the leftmost interval to infinity, while "r" means the same for the
rightmost interval. Thus, "lr" is frequently used to get indices that
are guaranteed to fall inside the table (provided that the table is at
least 2 elements long).

The common expression with the old lookup was to use
lookup(x(2:end-1), y)+1. This is dangerous, as it breaks for
decreasing tables of length 3
(see http://www.nabble.com/binary-lookup-tt16029897.html#a16089657)

I'm not sure whether this case can occur here, I've replaced all the
lookup(x(2:end-1)) expressions I could find since the new form is also
faster (no temporary arrays).


On Fri, Apr 25, 2008 at 9:17 AM, Shai Ayal <address@hidden> wrote:
> this is a trivial on-liner. I think lookup was changed recently. Since
>  I don't know the meaning of the "lr" parameter to lookup I tried to
>  guess.
>  This show up in making the doc/interpreter/interpderiv2.eps figure. It
>  looks right to me with the fix, but please double-check.
>
>  Shai
>
>
>  diff --git a/scripts/polynomial/ppval.m b/scripts/polynomial/ppval.m
>  --- a/scripts/polynomial/ppval.m
>  +++ b/scripts/polynomial/ppval.m
>  @@ -39,7 +39,7 @@ function yi = ppval (pp, xi)
>      transposed = (columns (xi) == 1);
>      xi = xi(:);
>      xn = length (xi);
>  -    idx = lookup (pp.x, xi, "lr");
>  +    idx = lookup (pp.x(2:end-1), xi)+1;
>      dx = (xi - pp.x(idx)).';
>      dx = reshape (dx(ones(1,prod(pp.d)),:),[pp.d,xn]);
>      c = reshape (pp.P(:,1), pp.n, prod (pp.d));
>



-- 
RNDr. Jaroslav Hajek
computing expert
Aeronautical Research and Test Institute (VZLU)
Prague, Czech Republic
url: www.highegg.matfyz.cz